James D. Shepherd

Correcting satellite imagery for the variance of reflectance and illumination with topography

Steephill and mountain slopes severely affect remote sensing of vegetation. The irradiation on a slope varies strongly with slope azimuth relative to the sun, and the reflectance of the slope varies with the angles of incidence and exitance relative to the slope normal. Topographic correction involves standardizing imagery for these two effects. We use an atmospheric model with a Digital Elevation Model (DEM) to calculate direct and diffuse illumination, and a simple function of incidence and exitance angles to calculate vegetation-canopy reflectance on terrain slope. The reflectance correction has been derived from the physics of visible direct radiation on a vegetation canopy, but has proved applicable to infrared wavelengths and only requires solar position, slope and aspect. We applied the reflectance and illumination correction to a SPOT 4 image of New Zealand to remove topographic variation. In all spectral bands, the algorithm markedly reduced the coefficients of variation of vegetation groups on rugged terrain. This produced clean spectral signatures, improving the capacity for automated classification. If illumination correction is performed alone, the coefficients of variation can be increased, and so should not be applied without a reflectance correction. The algorithm output is reflectance on a level surface, enabling the monitoring of vegetation in hilly and mountainous areas.

Correction of the topographic effect in remote sensing

We derive a formula for the dependence of vegetation-canopy reflectance on terrain slope (visible light only). Reflectance is inversely proportional to the sum of cosine of incidence angle and cosine of exitance angle. Laboratory measurements of miniature forest canopies set up on inclined slopes compare well with predicted reflectances.

BRDF Correction of Vegetation in AVHRR Imagery

In this article we present a method for correcting AVHRR visible and near-infrared imagery for varying satellite and solar zenith angles. This method is based on the WAK BRDF model for closed canopies. The parameters required to perform BRDF correction can be derived from consecutive pass AVHRR imagery pairs. This imagery provides two views of the land surface close together in time but with large differences in phase angle. It is reasonable to assume that both the surface and the atmosphere will change little between orbits, and that after BRDF correction reflectances of given targets should be the same in both orbits. Before BRDF parameters can be fitted, atmospheric correction must be performed. To improve this process, average monthly atmospheric profiles and aerosol optical depths are used as radiative transfer model inputs in conjunction with a digital elevation model. Using atmosphere corrected reflectance data from 12 NOAA-14 AVHRR image pairs, BRDF parameters were extracted for predominant vegetation groups in New Zealand: indigenous forest; exotic forest; scrub; pasture; and tussock grassland. For each of these vegetation groups significant non-Lambertian reflectance behavior was observed, and BRDF correction using the derived parameters successfully minimized this variation. Measuring the spread of the corrected results from the desired equal reflectance line gives a measure of the accuracy of the method. After correction, the RMS reflectance errors were approximately 0.01 in the visible and 0.02 in the near-infrared. A vegetation map specifying the proportions of the vegetation groups at any given location can be used to perform regular BRDF correction. Reflectance standardization to a fixed view and sun angle can then be performed using the pre-derived BRDF parameters and proportional BRDF correction.

A Python-Based Open Source System for Geographic Object-Based Image Analysis (GEOBIA) Utilizing Raster Attribute Tables

A modular system for performing Geographic Object-Based Image Analysis (GEOBIA), using entirely open source (General Public License compatible) software, is presented based around representing objects as raster clumps and storing attributes as a raster attribute table (RAT). The system utilizes a number of libraries, developed by the authors: The Remote Sensing and GIS Library (RSGISLib), the Raster I/O Simplification (RIOS) Python Library, the KEA image format and TuiView image viewer. All libraries are accessed through Python, providing a common interface on which to build processing chains. Three examples are presented, to demonstrate the capabilities of the system: (1) classification of mangrove extent and change in French Guiana; (2) a generic scheme for the classification of the UN-FAO land cover classification system (LCCS) and their subsequent translation to habitat categories; and (3) a national-scale segmentation for Australia. The system presented provides similar functionality to existing GEOBIA packages, but is more flexible, due to its modular environment, capable of handling complex classification processes and applying them to larger datasets.

Use of VEGETATION satellite imagery to map pasture quality for input to a methane budget of New Zealand

To calculate an accurate methane budget of New Zealand, it is necessary to measure the spatial and temporal variation of metabolizable energy of pasture. The VEGETATION sensor, on board SPOT4 and SPOT5, provides imagery at appropriate spatial and temporal scales. Imagery can be composited over 10 days to remove cloud cover and then processed to remove artefacts associated with directional reflectance. One year of VEGETATION imagery from March 2000 through to February 2001 was processed and co‐acquired with metabolizable energy measurements from 17 farms spread throughout New Zealand. Metabolizable energy was related to the Normalized Difference Vegetation Index (NDVI) with a Loess regression model to produce monthly maps of metabolizable energy. If these maps were used to estimate average metabolizable energy of pasture in a national methane budget, then the accuracy would not be increased (the uncertainty of average metabolizable energy from the satellite based method is the same as the nominal uncertainty assigned from expert judgement). However, the satellite‐based method would lend statistical credibility, and give the potential for spatial and temporal disaggregation of the budget.

Spatial prediction of brushtail possum (Trichosurus vulpecula) distribution using a combination of remotely sensed and field-observed environmental data

Abstract Context. In New Zealand, the introduced brushtail possum, Trichosurus vulpecula, is a reservoir of bovine tuberculosis and as such poses a major threat to the livestock industry. Aerial 1080 poisoning is an important tool for possum control but is expensive, creating an ongoing need for ever more cost-effective ways of using this technique. Aims. To develop geographic information system (GIS) models to better predict spatial variation in the distribution of unmanaged possum populations, to facilitate better targeting of control activities. Methods. Relative abundance of possums and their distribution among habitat types were surveyed in a dry high-country area of the northern South Island. Two GIS-based models were developed to predict the relative abundance of possums on trap lines. The first model used remotely sensed (digital) environmental data; the second complemented the remotely sensed data with fine-scale habitat and topographic data collected on the ground. Key results. Digital environmental factors and habitat features proved to be key predictors of relative possum abundance. In both GIS models, height above valley floor, presence of forest cover and mean annual temperature were the strongest predictors. Conclusions. Predictive maps (projections) of relative possum abundance produced from these models can provide useful decision-support tools for pest-control managers, by enabling possum control to be targeted spatially. Implications. Spatially targeted pest control could allow effective control activities for invasive species or disease vectors to be applied at a lower cost for the same benefit.

Optimizing the airborne thermal detection of possums

Brushtail possums (Trichosurus vulpecula

Accurate registration of optical satellite imagery with elevation models for topographic correction

It is necessary to remove the effects of topography from optical satellite imagery if automated techniques are to be used to infer surface properties. This is especially the case in mountainous terrain where variable slope normals cause variation in both illumination and reflectance of light. Digital elevation models (DEMs) are required to model slope normals and make topographic corrections. However, it is difficult to achieve accurate registration between ortho-rectified satellite images and DEMs. We show how this mis-registration, which can be spatially variable, may be accounted for with the use of a local correlation filter. The filter determines the offset between a DEM shade map and ortho-rectified satellite image for every pixel. Association of satellite image pixels with the ‘correct’ slope normal in topographic correction removes the majority of ghosting and high-frequency artefacts.

Support vector machine classification of woody patches in New Zealand from synthetic aperture radar and optical data, with LiDAR training

Abstract. A light detection and ranging canopy height model (CHM) was used as training data for a segment-based classification of woody patches. The classifier is accurate (∼92%) and suitable for use at the national scale. Height thresholds and percentage cover of vegetation from the CHM were used to produce larger quantities of reliable training data compared to other, mostly point or plot-based, ground-truthing approaches. It was found that the regional-scale differentiation between woody and nonwoody vegetation might be achieved by a combination of L-band dual-polarized Phased Array type L-band synthetic aperture radar data (HV) with multispectral optical data that include a short-wave infrared band. The application of a support vector machine algorithm to these data proved successful. The versatility of these algorithms regarding the discrimination function and their ability to solve classification problems with multiple output classes were critical factors for success. The identified and classified woody patches constitute a valuable addition and enhancement of the national land cover database.

Multi-scale dynamic maps for the management of invading and established wildlife populations: brushtail possums in New Zealand

Abstract Context. The abundance and distribution of mammalian species often change in response to environmental variability, losses or gains in suitable habitat and, in the case of pest species, control programs. Consequently, conventional distribution maps rapidly become out of date and fail to provide useful information for wildlife managers. For invasive brushtail possum populations in New Zealand, the main causes of change are control programs by central and local government agencies, and post-control recovery through recolonisation and in situ recruitment. Managers need to know current, and likely future, possum population levels relative to control targets to help assess success at preventing the spread of disease or for protecting indigenous species. Information on the outcomes of government-funded possum control needs to be readily available to members of the general public interested in issues such as conservation, disease management and animal welfare. Aims. To produce dynamic, scalable maps of the current and predicted future distribution and abundance of possums in New Zealand, taking into account changes due to control, and to use recent visualisation technology to make this information accessible to managers and the general public for assessing control strategies at multiple spatial scales. Methods. We updated an existing individual-based spatial model of possum population dynamics, extending it to represent all individuals in a national population of up to 40 million. In addition, we created a prototype interface for interactive web-based presentation of the model’s predictions. Key results. The improved capability of the new model for assessing possum management at local-to-national scales provided for real-time, mapped updates and forecasts of the distribution and abundance of possums in New Zealand. The versatility of this platform was illustrated using scenarios for current and emerging issues in New Zealand. These are hypothetical incursions of possums, reinvasion of large areas cleared of possums, and impacts on animal welfare of national-scale management of possums as vectors of bovine tuberculosis (TB). Conclusions. The new individual-based spatial model for possum populations in New Zealand demonstrated the utility of combining models of wildlife population dynamics with high-speed computing capability to provide up-to-date, easily accessible information on a species’ distribution and abundance. Applications include predictions for future changes in response to incursions, reinvasion and large-scale possum control. Similar models can be used for other species for which there are suitable demographic data, typically pest species, harvested species or species with a high conservation value. Implications. Models such as the spatial model for possums in New Zealand can provide platforms for (1) real-time visualisation of wildlife distribution and abundance, (2) reporting and assessing progress towards achieving management goals at multiple scales, (3) use as a decision-support tool to scope potential changes in wildlife populations or simulate the outcomes of alternative management strategies, and (4) making information about pest control publicly available.